Summary of the invention
For realizing the correct identification of operator's wrist translational speed and motor pattern in Teleoperation Systems, the present invention proposes the electromyographic signal recognition methods of a kind of L-Z complexity and fractal dimension and maximum fractal length.First from related muscles group, gather corresponding surface electromyogram signal, then extract the L-Z complexity of electromyographic signal and fractal dimension as proper vector, finally take L-Z complexity and fractal dimension as proper vector input K arest neighbors model Incremental Learning Algorithm sorter, realize the upper limbs multi-pattern recognition of electromyographic signal, extract the fractal length of maximum of electromyographic signal as controlled quentity controlled variable, the speed of identification upper extremity exercise.
In order to realize above object, the inventive method mainly comprises the following steps:
Step (1). obtain human upper limb electromyographic signal sample data, specifically: first by electromyographic signal collection instrument, pick up human upper limb electromyographic signal, then adopt the signal noise silencing method based on Wavelet Energy Spectrum entropy to carry out de-noising to the electromyographic signal that contains interference noise.
Step (2). the electromyographic signal that step (1) is obtained is carried out feature extraction, obtains L-Z complexity, fractal dimension and the maximum fractal length of this electromyographic signal.
Described L-Z complexity, specific algorithm is as follows:
Lempel-Ziv complexity is proposed by Lempel and Ziv a kind of for describing the nonlinear indicator of the random degree of sequence, the algorithm that the calculating of its value c (n) adopts Kaspar and Schusyer to propose conventionally.
The symbolism sequence of supposing original signal is s
1s
2... s
n.From empty string, start to add s
1, by copying and adding, operate the connection that realizes complete sequence.If generated prefix s
1s
2∧ s
r-1, r < n, and next symbol s
rwith adding, operated, be designated as:
s
1s
2∧s
r-1→s
1s
2∧s
r-1s
r· (1)
Here at s
rafter mark " " reflected s
rgenerative process-interpolation.Be the detailed process realizing below:
Make S=s
1s
2... s
r, Q=s
r+1, SQ represents S, total character string that Q is spliced into, and whether SQ π represents last character in SQ to leave out the character string of gained, observe Q and can with clone method, obtain from certain symbol of SQ π.If Q can not be from SQ π certain substring copy and obtain, just with adding to operate, add s
r+1, and marking " ".If Q can be from SQ π certain symbol copy and obtain, continuing to observe increases character s
r+2s
r+3string, check whether reproducible obtains for it, until can not, do and add operation, and marking " ", repeat said process to last character s of sequence
n.In sequence, the number m of mark " " has reflected the number of times of taking to add operation." complexity " c (n) can be tried to achieve by following formula:
c(n)=(m.log
2n)/n (2)
Wherein, m is for adding the number of times of operation, the length that n is sequence.This Complexity Measurement c (n) of Kaspar-Schuster is simple to operate, is easy to realize.
Described fractal dimension and maximum fractal length, specific algorithm is as follows:
Fractal dimension calculates the method that adopts the Higuchi T of Tokyo Univ Japan professor to propose.The time series of supposing equal interval sampling and obtaining is: X=(x
1, x
2..., x
n), constructor sequence set
In formula, under int representative, round, k, m are integer.To appointing, get an integer k and can obtain k group sequence.
In formula, N-1/ (int ((N-m)/k) .k) is the normalized factor of Time Sub-series length, and is defined as L corresponding to length L EssT.LTssT.LT L (k) > of k
m(k) average.As L (k)=Ck
-D, D is seasonal effect in time series fractal dimension, and natural logarithm is got in equation left and right, has
ln L(k)=ln C-D ln k (4)
Analyze knownly, ln k and ln L (k) are that slope is-linear relation of D, if can try to achieve ln k and ln L (k), use these points of least square fitting, can obtain Fractal dimensions.
The Naik of Univ Melbourne Australia, length L EssT.LTssT.LT L (k) > during by smallest dimension k such as Ganesh R is defined as maximum fractal length (maximum fractal length MFL).
Step (3). the L-Z complexity that the step (2) of usining is obtained and fractal dimension, as proper vector input K arest neighbors model Incremental Learning Algorithm sorter, obtain recognition result.
The present invention proposes and develop on a kind of K arest neighbors (kNN) method basis, adopt the improved K arest neighbors of clustering technique model Incremental Learning Algorithm.
The data preparation of this algorithm, adopts C-means clustering algorithm, usings central point as the representative point of KNN algorithm, and algorithm steps is described below.
(1) use C-means clustering algorithm respectively the sample point in each classification to be carried out to automatic cluster, suppose that the sample in each classification is polymerized to respectively m bunch.
(2) all samples in same cluster are simply calculated, set up the four-tuple of model.
In the application of multimode recognition, correct pattern-recognition result should can be used as follow-up pattern-recognition foundation, and therefore, the model that algorithm should possess in adjusting and improve bunch is learnt and receives newly-increased sample data, reaches the effect of incremental learning.For the sample point being verified through pattern-recognition, when number does not also reach a certain amount of accumulation, adopt the similar criterion of C-mean cluster to be integrated in some bunch of class, and to bunch four-tuple data adjust.When the sample points after checking reaches a certain amount of accumulation, use again C-means clustering algorithm cluster again, set up new model cluster.
(3) by K arest neighbors (kNN) method, calculate the distance of sample to be sorted and each representative point (central point).When choosing the classification of sample to be identified in a vote, each is represented to dot product one weight coefficient
by the data ballot after weighting, determined the classification of sample to be identified.
Step (4). the fractal length of maximum of the electromyographic signal that the step (2) of usining is obtained, as controlled quentity controlled variable, realizes the control that mechanical arm captures speed.
In order to make the motion of mechanical arm keep synchronizeing with the motion of the main hand of operator as far as possible, strengthen telepresenc, the sense of reality of distant operation, the present invention controls and designs mechanical arm crawl speed.According to the responsiveness of surface electromyogram signal characteristic parameter quantified controlling robot, signal is stronger, and characteristic ginseng value is larger, move faster, otherwise slow.The fractal length of maximum of electromyographic signal is to characterize the suitable characteristic parameter of muscle activity intensity, and is convenient to calculate, and it is convenient to realize, and the size of its value and grasp speed are monotone increasing relation.Thereby, be to control desirable reference input.Be subject to the restriction of mechanical hardware condition, main operation person's manual manipulation velocity range is taken as 0.1~0.5 (1/s).If with m=av
2+ bv+c represents the relation between operating speed and the maximum fractal length of flesh signal, and wherein: v is speed, m is maximum fractal length, and the funtcional relationship on extensor and musculus flexor is as follows:
Extensor:
Musculus flexor:
The size of manipulator control input quantity is determined by the maximum fractal length weighted sum of the electromyographic signal on extensor and musculus flexor:
M
x=aM
1+bM
2 (8)
Wherein, M
xfor weighing the fractal length of total maximum of proportional control factor, M in certain time period
1for the fractal length of maximum of extensor in this time period, M
2for the fractal length of maximum of musculus flexor in this time period, a, b is respectively weighted value.
The value of [a, b] is relevant with pattern.As stretched under wrist pattern, directly related with this pattern is extensor, and the impact of musculus flexor is quite faint, therefore [a, b] desirable [1,0]; Under wrist pattern in the wrong, directly related with this pattern is musculus flexor, and the impact of extensor is quite faint, therefore [a, b] desirable [0,1]; What the present invention discussed is grasping movement pattern, and directly related with this pattern is extensor and musculus flexor, therefore [a, b] desirable [1/2,1/2].
The present invention compares with existing many hand muscle electric signal action identification methods, has following features:
Utilize the L-Z complexity index of flesh signal and the identification that the fractal dimension feature in fractal theory realizes Wrist-sport pattern, result is controlled the pattern of floor-washing robot mechanical arm far away for operator.Movement recognition sorter has adopted the improved KNN model of C-means clustering technology Incremental Learning Algorithm, not only inherited the advantage that KNN algorithm performance is stable, discrimination is high, and possessed the ability of incremental learning, along with the increase of pattern-recognition sample, discrimination can further improve.The responsiveness of hand depends on the activity intensity of arm muscle group, muscle activity intensity can fractal length be characterized by the maximum of electromyographic signal, therefore the fractal length of maximum of electromyographic signal of take is input control amount, realized the control that mechanical arm captures speed, obtained comparatively ideal effect.
Embodiment
Below in conjunction with accompanying drawing, embodiments of the invention are elaborated: the present embodiment is implemented take technical solution of the present invention under prerequisite, has provided detailed embodiment and concrete operating process.
As shown in Figure 1, the present embodiment comprises the steps:
Step 1, obtains human upper limb electromyographic signal sample data, specifically: first by electromyographic signal collection instrument, pick up human upper limb electromyographic signal, then adopt the signal noise silencing method based on Wavelet Energy Spectrum entropy to carry out de-noising to the electromyographic signal that contains interference noise.
(1) gather the electromyographic signal of human upper limb.Experimenter clenches fist respectively, each 50 groups of exrending boxing, wrist inward turning and 4 actions of wrist outward turning, selects upper limbs musculus extensor carpi ulnaris and musculus flexor carpi ulnaris to originate as surface electromyogram signal.During experiment, first with alcohol obliterating decontamination on experimenter's musculus extensor carpi ulnaris and musculus flexor carpi ulnaris respectively, to strengthen picking up signal ability, adopt MyoTrace400 electromyographic signal collection instrument to pick up musculus extensor carpi ulnaris and surface electromyogram signal corresponding to musculus flexor carpi ulnaris.
(2) adopt the signal noise silencing method based on Wavelet Energy Spectrum entropy to carry out de-noising to the electromyographic signal that contains interference noise.
Step 2, the electromyographic signal that step 1 is obtained is carried out feature extraction, obtains L-Z complexity, fractal dimension and the maximum fractal length of this electromyographic signal.
Seasonal effect in time series coarse is prerequisite and the key of calculating kmpel-ziv complexity.The present invention adopts a kind of multiple dimensioned coarse method; adopt four intervals; with two scale-of-two, former electromyographic signal is encoded; the first is rough segmentation position; " 1 " represents that signal is greater than mean value, " 0 " otherwise, second is refinement position; " 1 " represents that signal is in interval top, and " 0 " represents in interval bottom.Like this, the reconstruct of sequence is no longer just binaryzation, but many-valued coarse algorithm represents with " 0~3 ".Adopting this its advantage of coarse method is that original signal and reproducing sequence are man-to-man mapping relations, and the reproducing sequence of signal shown in Fig. 2 becomes ((23131020).
Electromyographic signal complexity algorithmic procedure is as follows:
1) ask local maximum, the minimal value of electromyographic signal.By the envelope up and down of interpolating function picked up signal.And upper and lower envelope is averaging, be designated as m (i).To substitute the original value of electromyographic signal x (i) in corresponding sequence of points with value of symbol s (i) on envelope, ask the difference of signal s (i) and m (i), be designated as h (i)=s (i)-m (i).Above, the ordinal number that i is electromyographic signal, its value is 1 to N, the length that N is signal.
2) h (i) signal normalization: the amount h that finds out absolute value maximum in h (i)
max=max|h (i) | i=1,2 ... N.Finally, by amplitude normalization:
EMG(i)=h(i)/h
max i=1,2,...,N
3) time series signal is carried out to the multiple dimensioned many-valued coarse of four-range.
4) with Kaspar-Schusyer method, calculate the complexity of signal.
In asking the algorithm of electromyographic signal complexity, it is that method is successfully crucial that the 1st step replaces original signal value with the value on envelope.Reason is, identical pattern, and the fluctuation tendency of electromyographic signal is identical, and movement velocity of details and hand, dynamics size etc. are relevant, as directly asked for the complexity of signal with original signal, the value mobility scale of same pattern complexity can be larger.For pattern-recognition, do not reach desirable effect.
Table 1 is for respectively getting the statistics of 50 groups of electromyographic signals electromyographic signal complexity on musculus extensor carpi ulnaris and musculus flexor carpi ulnaris to four class patterns.
Table 1 surface electromyogram signal complexity index statistics
At the fractal dimension for pattern-recognition, calculate, the sampling number of electromyographic signal is got 2000, and the value of k gets respectively 2
0, 2
1..., 2
8.Table 2 is the statistics to 50 groups of electromyographic signals of four class patterns its fractal dimension of electromyographic signal on musculus extensor carpi ulnaris and musculus flexor carpi ulnaris.
Table 2 surface electromyogram signal dimension statistics
The activity intensity of the numerical value of gained < L (k) > related muscles when also the fractal length of maximum of electromyographic signal can reflect operating state well during small scale k.The maximum fractal length value relation of responsiveness and related muscles group electromyographic signal under measurement operation hand grasp mode.The curve that on the grasp speed that Fig. 3 (a) and (b) obtain while being respectively experiment and extensor, musculus flexor, the fractal length respective value of maximum of electromyographic signal is described, horizontal ordinate adopts normalized velocity, also be operator's hand from the straight inverse starting to holding the spherical object time used that opens, unit is 1/s.Ordinate is the fractal length of maximum (MFL) average of electromyographic signal on extensor and musculus flexor while being converted into 2000 sampled points, immeasurable just.As can be seen from the figure, the fractal length of the maximum of muscle groups electromyographic signal is monotone increasing along with the increase of grasp speed, and speed remains unchanged after reaching 0.6 (1/s) substantially.
Step 3. the L-Z complexity that the step 2 of usining is obtained and fractal dimension, as proper vector input K arest neighbors model Incremental Learning Algorithm sorter, obtain recognition result.
When tele-robotic system is just built sorter, first calculate complexity and the fractal dimension of electromyographic signal on two groups of muscle, form four-dimensional vectorial sample.The m value of C-means clustering algorithm gets 15, carries out the arrangement of sample points certificate, then follow-through is carried out to pattern-recognition.
Table 3 is the service condition of sorter after setting up.First is for just building the pattern-recognition results to follow-up 50 groups of actions after sorter.Second portion is added up for moving by every class the discrimination that continues identification after 50 groups of incremental learnings again.Result shows, and accumulates by a certain amount of incremental learning, and discrimination has further raising, and effect is also comparatively desirable.
Table 3 sorter recognition result
Step 4. the fractal length of maximum of the electromyographic signal that the step 2 of usining is obtained, as controlled quentity controlled variable, realizes the control that mechanical arm captures speed.
The maximum fractal length of the electromyographic signal of obtaining is converted into the form of PWM dutycycle by a kind of linear corresponding relation, to reach the responsiveness of controlling direct current generator.Fig. 4 is that the designed remote control system main operation person hand of the present invention and mechanical arm capture time consistency contrast test data and curves, and horizontal ordinate represents main operation person's grasp speed, and ordinate represents the mechanical arm relative error that puts in place.Error is in the patient scope of main operation person.